Nearly three hours with a cable company is enough time to develop a philosophy of customer service. Gustavo Sapoznik developed a business. He had been looking for a problem big enough to justify building an artificial intelligence company. Then one arrived through the telephone, accompanied by the familiar indignities of trying to get something ordinary done.
The interesting part came when he considered the bill for the experience. His time had been wasted. Someone on the other end had been paid to waste it with him. Here was an arrangement that could leave both customer and company poorer, without producing much satisfaction for either. An expensive machine was turning, and its output was frustration.
Sapoznik eventually changed cable providers. That small act matters to the larger story. A customer-service interaction can look like an expense on a spreadsheet, yet it can also decide whether the next payment arrives at all. He began looking at contact centers as places where the economics of a business and the patience of its customers meet, often under rather poor lighting.
The view from six billing systems
There is a second scene in Sapoznik's account of the industry that deserves equal billing. While shadowing a contact-center worker at a large company, he watched the agent enter a caller's information into six billing systems before finding it. The customer's problem had become a search expedition inside the company's own software.
He called the assortment a “Frankenstack.” It is an efficient nickname for an inefficient arrangement. Each program may have a perfectly respectable purpose. Together, they can require the person using them to remember where everything lives while keeping a conversation going with somebody who would quite reasonably prefer an answer.
Consider what the job asks of that worker: listen, interpret, search, remember, type, comply with policy, and sound composed. The caller hears pauses. The employee sees another window. A business that measures only the duration of the call can miss the little obstacle course that determines its duration.
That encounter gives Sapoznik's career its human scale. ASAPP's subject is artificial intelligence, but the recurring character in its early story is an employee juggling systems. The technology has to earn its place by reducing that juggling. A fluent sentence is useful; getting the account information onto the right screen is useful too.
“That was an eye-opening moment.”Gustavo Sapoznik, on shadowing a contact-center agent
A complaint with selection criteria
Sapoznik dates ASAPP's beginnings to 2014. Before settling on customer service, he wanted a business problem with three properties: substantial economic scale, persistent inefficiency, and abundant data. The cable call provided a rather persuasive demonstration of the middle condition. The industry's spending and constant stream of conversations supplied the others.
Those criteria are revealing because they put the problem ahead of the product demonstration. If an organization already works well, new software has to fight for a marginal improvement. If its work generates little usable information, a learning system has little to learn from. Customer service offered both accumulated knowledge and visible room for improvement.
The idea was to make the people already doing the work more productive. In its early customer-service applications, ASAPP suggested responses during live conversations, brought relevant information forward, and helped with the administrative work around an interaction. For messaging agents, a proposed answer could become a click rather than another passage typed from scratch.
The difference sounds modest at the level of one message. Across repeated interactions, however, the boring details become the business. Every piece of information retrieved twice, every answer rewritten, and every summary composed after a call consumes attention that could have gone into the next customer's problem.
A schematic of the problem Sapoznik pursued: a conversation is only part of a service interaction.
The research department gets the chairs
Sapoznik built ASAPP with research and engineering close to its center. In 2020, he described nearly 80 percent of the team as belonging to research and development. The work spanned machine learning, natural language processing, and speech recognition. These were disciplines applied to a workplace whose frustrations were already well established.
His family connections also appeared in the company's geography. Sapoznik is the son of Argentine parents. ASAPP established a presence in Buenos Aires in late 2018, recruiting scientists, mathematicians, and technical specialists. The company's New York identity did not confine its search for people who could tackle the underlying problems.
By October 2022, ASAPP had announced an engineering office in Bengaluru, alongside offices in New York, Mountain View, Bozeman, and Buenos Aires. Its technical organization then included 25 PhDs. The company reported 35 approved patents and another 50 pending. Those figures describe an investment in building technology, rather than simply giving an existing service a new label.
The products coming out of that organization addressed different pieces of an agent's day: transcription, suggested replies, and conversation summaries. A summary may lack the theatrical appeal of a talking computer. Anyone who has finished a long conversation and still had to write it up can appreciate its charms.
There is a practical discipline in that choice of work. Customer service asks software to function amid accents, interruptions, incomplete requests, and company-specific procedures. The research has to survive contact with those conditions. An elegant model that adds another task to the agent's list would be an awkward answer to the problem Sapoznik had noticed.
The money follows the minutes
ASAPP's financing made the cost of pursuing that ambition visible. In May 2020, it announced a $185 million Series B and $260 million in total investment. Sapoznik described the approach as improving agents' capabilities and automating their workflows. The objective was to change how the work got done inside large enterprises.
A year later, on May 19, 2021, the company announced a $120 million Series C, bringing its stated total funding to $400 million. The round valued ASAPP at $1.6 billion, twice the valuation at its previous financing. Fidelity Management & Research Company and Dragoneer joined the investor group.
JetBlue and DISH were among the customers named during this period. These businesses gave the proposition a concrete setting: travel arrangements, accounts, and service questions that arrive in volume. The significance of a faster response depends on the work behind it and on how often a company has to perform that work.
A valuation is an investor's judgment at a particular moment. The daily test of the business remains less glamorous. Can the employee answer correctly? Can the customer finish what they came to do? Can the company serve that customer without spending nearly three hours on an exchange neither party enjoyed?
May 2021
at that financing
A dated financing snapshot. These figures describe ASAPP, rather than Sapoznik's personal wealth.
Experienced company, uncomfortable questions
Sapoznik gathered people around ASAPP who had spent years building and financing enterprise technology businesses. John Doerr became an investor and board member. John Chambers, the former Cisco chief executive, became an advisor. Those connections brought experience with the large organizations ASAPP hoped to persuade.
In September 2023, Frank Slootman, then Snowflake's chief executive and chairman, joined the board. His earlier leadership roles included ServiceNow and Data Domain. Sapoznik welcomed his experience as ASAPP worked on scaling the business. A company can build convincing technology and still need considerable help turning it into a repeatable commercial operation.
The gathering of experienced operators fits a less romantic part of the founder's job: an enterprise customer has procurement processes, existing systems, and consequences for getting service wrong. Winning that customer requires more than enthusiasm. It requires an organization that can implement what it sells and support it after the presentation ends.
From suggested replies to completed work
The emphasis evolved as the technology did. In April 2024, ASAPP announced GenerativeAgent, a conversational voice and messaging product designed to automate customer-service interactions. The launch moved the discussion beyond helping an employee compose the next answer toward having software perform more of the interaction itself.
That is a consequential expansion of the original proposition. An assistant can offer a suggestion for a person to inspect. A system taking action needs access to information, rules about what it may do, and a way to bring a person into the process when necessary. The quality of the conversation remains tied to what happens in the company's systems.
Sapoznik's ambition also extended to a more proactive relationship between customer and business. In a 2023 conversation with Daniel Darling, he described a future in which companies anticipate problems instead of waiting for a customer to report them. A service organization could contact somebody because it had recognized a difficulty in advance.
The appeal is easy to understand. Nobody chooses to spend an afternoon explaining an avoidable problem. But that aspiration sets a demanding standard: the contact must be timely, relevant, and capable of helping. An unsolicited message that merely announces a new inconvenience would give the customer another reason to call.
The call outlasts the job title
Sapoznik's career has since moved into another chapter. His current public profile identifies him as founder and CEO of Conquest, based in Bozeman, Montana. ASAPP lists him as founder and advisor. Katie Stein is now its chief executive. His place in the company's story is therefore both foundational and distinct from its current operating leadership.
The ASAPP years leave a clear account of what caught his attention: the gap between what an organization knows and what a customer can get it to do. He approached that gap through research, product development, enterprise customers, and an expensive commitment to technical talent. The ordinary telephone call provided an unusually durable question.
Nearly three hours is a poor service experience and a generous amount of thinking time. Sapoznik used it to consider the person taking the call, the company paying for it, and the systems keeping them occupied. The opportunity lay in making an everyday exchange less laborious. There are few prizes more readily understood by a customer than getting the afternoon back.